1 /* 2 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved. 3 * 4 * Authors: 5 * Alexander Graf <agraf@suse.de> 6 * Kevin Wolf <mail@kevin-wolf.de> 7 * Paul Mackerras <paulus@samba.org> 8 * 9 * Description: 10 * Functions relating to running KVM on Book 3S processors where 11 * we don't have access to hypervisor mode, and we run the guest 12 * in problem state (user mode). 13 * 14 * This file is derived from arch/powerpc/kvm/44x.c, 15 * by Hollis Blanchard <hollisb@us.ibm.com>. 16 * 17 * This program is free software; you can redistribute it and/or modify 18 * it under the terms of the GNU General Public License, version 2, as 19 * published by the Free Software Foundation. 20 */ 21 22 #include <linux/kvm_host.h> 23 #include <linux/export.h> 24 #include <linux/err.h> 25 #include <linux/slab.h> 26 27 #include <asm/reg.h> 28 #include <asm/cputable.h> 29 #include <asm/cacheflush.h> 30 #include <asm/tlbflush.h> 31 #include <asm/uaccess.h> 32 #include <asm/io.h> 33 #include <asm/kvm_ppc.h> 34 #include <asm/kvm_book3s.h> 35 #include <asm/mmu_context.h> 36 #include <asm/switch_to.h> 37 #include <linux/gfp.h> 38 #include <linux/sched.h> 39 #include <linux/vmalloc.h> 40 #include <linux/highmem.h> 41 42 #include "trace.h" 43 44 /* #define EXIT_DEBUG */ 45 /* #define DEBUG_EXT */ 46 47 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr, 48 ulong msr); 49 50 /* Some compatibility defines */ 51 #ifdef CONFIG_PPC_BOOK3S_32 52 #define MSR_USER32 MSR_USER 53 #define MSR_USER64 MSR_USER 54 #define HW_PAGE_SIZE PAGE_SIZE 55 #define __hard_irq_disable local_irq_disable 56 #define __hard_irq_enable local_irq_enable 57 #endif 58 59 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu) 60 { 61 #ifdef CONFIG_PPC_BOOK3S_64 62 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu); 63 memcpy(svcpu->slb, to_book3s(vcpu)->slb_shadow, sizeof(svcpu->slb)); 64 memcpy(&get_paca()->shadow_vcpu, to_book3s(vcpu)->shadow_vcpu, 65 sizeof(get_paca()->shadow_vcpu)); 66 svcpu->slb_max = to_book3s(vcpu)->slb_shadow_max; 67 svcpu_put(svcpu); 68 #endif 69 70 #ifdef CONFIG_PPC_BOOK3S_32 71 current->thread.kvm_shadow_vcpu = to_book3s(vcpu)->shadow_vcpu; 72 #endif 73 } 74 75 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu) 76 { 77 #ifdef CONFIG_PPC_BOOK3S_64 78 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu); 79 memcpy(to_book3s(vcpu)->slb_shadow, svcpu->slb, sizeof(svcpu->slb)); 80 memcpy(to_book3s(vcpu)->shadow_vcpu, &get_paca()->shadow_vcpu, 81 sizeof(get_paca()->shadow_vcpu)); 82 to_book3s(vcpu)->slb_shadow_max = svcpu->slb_max; 83 svcpu_put(svcpu); 84 #endif 85 86 kvmppc_giveup_ext(vcpu, MSR_FP); 87 kvmppc_giveup_ext(vcpu, MSR_VEC); 88 kvmppc_giveup_ext(vcpu, MSR_VSX); 89 } 90 91 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu) 92 { 93 ulong smsr = vcpu->arch.shared->msr; 94 95 /* Guest MSR values */ 96 smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_DE; 97 /* Process MSR values */ 98 smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE; 99 /* External providers the guest reserved */ 100 smsr |= (vcpu->arch.shared->msr & vcpu->arch.guest_owned_ext); 101 /* 64-bit Process MSR values */ 102 #ifdef CONFIG_PPC_BOOK3S_64 103 smsr |= MSR_ISF | MSR_HV; 104 #endif 105 vcpu->arch.shadow_msr = smsr; 106 } 107 108 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr) 109 { 110 ulong old_msr = vcpu->arch.shared->msr; 111 112 #ifdef EXIT_DEBUG 113 printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr); 114 #endif 115 116 msr &= to_book3s(vcpu)->msr_mask; 117 vcpu->arch.shared->msr = msr; 118 kvmppc_recalc_shadow_msr(vcpu); 119 120 if (msr & MSR_POW) { 121 if (!vcpu->arch.pending_exceptions) { 122 kvm_vcpu_block(vcpu); 123 vcpu->stat.halt_wakeup++; 124 125 /* Unset POW bit after we woke up */ 126 msr &= ~MSR_POW; 127 vcpu->arch.shared->msr = msr; 128 } 129 } 130 131 if ((vcpu->arch.shared->msr & (MSR_PR|MSR_IR|MSR_DR)) != 132 (old_msr & (MSR_PR|MSR_IR|MSR_DR))) { 133 kvmppc_mmu_flush_segments(vcpu); 134 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)); 135 136 /* Preload magic page segment when in kernel mode */ 137 if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) { 138 struct kvm_vcpu_arch *a = &vcpu->arch; 139 140 if (msr & MSR_DR) 141 kvmppc_mmu_map_segment(vcpu, a->magic_page_ea); 142 else 143 kvmppc_mmu_map_segment(vcpu, a->magic_page_pa); 144 } 145 } 146 147 /* Preload FPU if it's enabled */ 148 if (vcpu->arch.shared->msr & MSR_FP) 149 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP); 150 } 151 152 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr) 153 { 154 u32 host_pvr; 155 156 vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB; 157 vcpu->arch.pvr = pvr; 158 #ifdef CONFIG_PPC_BOOK3S_64 159 if ((pvr >= 0x330000) && (pvr < 0x70330000)) { 160 kvmppc_mmu_book3s_64_init(vcpu); 161 if (!to_book3s(vcpu)->hior_explicit) 162 to_book3s(vcpu)->hior = 0xfff00000; 163 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL; 164 vcpu->arch.cpu_type = KVM_CPU_3S_64; 165 } else 166 #endif 167 { 168 kvmppc_mmu_book3s_32_init(vcpu); 169 if (!to_book3s(vcpu)->hior_explicit) 170 to_book3s(vcpu)->hior = 0; 171 to_book3s(vcpu)->msr_mask = 0xffffffffULL; 172 vcpu->arch.cpu_type = KVM_CPU_3S_32; 173 } 174 175 kvmppc_sanity_check(vcpu); 176 177 /* If we are in hypervisor level on 970, we can tell the CPU to 178 * treat DCBZ as 32 bytes store */ 179 vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32; 180 if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) && 181 !strcmp(cur_cpu_spec->platform, "ppc970")) 182 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32; 183 184 /* Cell performs badly if MSR_FEx are set. So let's hope nobody 185 really needs them in a VM on Cell and force disable them. */ 186 if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be")) 187 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1); 188 189 #ifdef CONFIG_PPC_BOOK3S_32 190 /* 32 bit Book3S always has 32 byte dcbz */ 191 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32; 192 #endif 193 194 /* On some CPUs we can execute paired single operations natively */ 195 asm ( "mfpvr %0" : "=r"(host_pvr)); 196 switch (host_pvr) { 197 case 0x00080200: /* lonestar 2.0 */ 198 case 0x00088202: /* lonestar 2.2 */ 199 case 0x70000100: /* gekko 1.0 */ 200 case 0x00080100: /* gekko 2.0 */ 201 case 0x00083203: /* gekko 2.3a */ 202 case 0x00083213: /* gekko 2.3b */ 203 case 0x00083204: /* gekko 2.4 */ 204 case 0x00083214: /* gekko 2.4e (8SE) - retail HW2 */ 205 case 0x00087200: /* broadway */ 206 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS; 207 /* Enable HID2.PSE - in case we need it later */ 208 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29)); 209 } 210 } 211 212 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To 213 * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to 214 * emulate 32 bytes dcbz length. 215 * 216 * The Book3s_64 inventors also realized this case and implemented a special bit 217 * in the HID5 register, which is a hypervisor ressource. Thus we can't use it. 218 * 219 * My approach here is to patch the dcbz instruction on executing pages. 220 */ 221 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte) 222 { 223 struct page *hpage; 224 u64 hpage_offset; 225 u32 *page; 226 int i; 227 228 hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT); 229 if (is_error_page(hpage)) { 230 kvm_release_page_clean(hpage); 231 return; 232 } 233 234 hpage_offset = pte->raddr & ~PAGE_MASK; 235 hpage_offset &= ~0xFFFULL; 236 hpage_offset /= 4; 237 238 get_page(hpage); 239 page = kmap_atomic(hpage); 240 241 /* patch dcbz into reserved instruction, so we trap */ 242 for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++) 243 if ((page[i] & 0xff0007ff) == INS_DCBZ) 244 page[i] &= 0xfffffff7; 245 246 kunmap_atomic(page); 247 put_page(hpage); 248 } 249 250 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) 251 { 252 ulong mp_pa = vcpu->arch.magic_page_pa; 253 254 if (unlikely(mp_pa) && 255 unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) { 256 return 1; 257 } 258 259 return kvm_is_visible_gfn(vcpu->kvm, gfn); 260 } 261 262 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu, 263 ulong eaddr, int vec) 264 { 265 bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE); 266 int r = RESUME_GUEST; 267 int relocated; 268 int page_found = 0; 269 struct kvmppc_pte pte; 270 bool is_mmio = false; 271 bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false; 272 bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false; 273 u64 vsid; 274 275 relocated = data ? dr : ir; 276 277 /* Resolve real address if translation turned on */ 278 if (relocated) { 279 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data); 280 } else { 281 pte.may_execute = true; 282 pte.may_read = true; 283 pte.may_write = true; 284 pte.raddr = eaddr & KVM_PAM; 285 pte.eaddr = eaddr; 286 pte.vpage = eaddr >> 12; 287 } 288 289 switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) { 290 case 0: 291 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12)); 292 break; 293 case MSR_DR: 294 case MSR_IR: 295 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid); 296 297 if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR) 298 pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12)); 299 else 300 pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12)); 301 pte.vpage |= vsid; 302 303 if (vsid == -1) 304 page_found = -EINVAL; 305 break; 306 } 307 308 if (vcpu->arch.mmu.is_dcbz32(vcpu) && 309 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) { 310 /* 311 * If we do the dcbz hack, we have to NX on every execution, 312 * so we can patch the executing code. This renders our guest 313 * NX-less. 314 */ 315 pte.may_execute = !data; 316 } 317 318 if (page_found == -ENOENT) { 319 /* Page not found in guest PTE entries */ 320 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu); 321 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu); 322 vcpu->arch.shared->dsisr = svcpu->fault_dsisr; 323 vcpu->arch.shared->msr |= 324 (svcpu->shadow_srr1 & 0x00000000f8000000ULL); 325 svcpu_put(svcpu); 326 kvmppc_book3s_queue_irqprio(vcpu, vec); 327 } else if (page_found == -EPERM) { 328 /* Storage protection */ 329 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu); 330 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu); 331 vcpu->arch.shared->dsisr = svcpu->fault_dsisr & ~DSISR_NOHPTE; 332 vcpu->arch.shared->dsisr |= DSISR_PROTFAULT; 333 vcpu->arch.shared->msr |= 334 svcpu->shadow_srr1 & 0x00000000f8000000ULL; 335 svcpu_put(svcpu); 336 kvmppc_book3s_queue_irqprio(vcpu, vec); 337 } else if (page_found == -EINVAL) { 338 /* Page not found in guest SLB */ 339 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu); 340 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80); 341 } else if (!is_mmio && 342 kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) { 343 /* The guest's PTE is not mapped yet. Map on the host */ 344 kvmppc_mmu_map_page(vcpu, &pte); 345 if (data) 346 vcpu->stat.sp_storage++; 347 else if (vcpu->arch.mmu.is_dcbz32(vcpu) && 348 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) 349 kvmppc_patch_dcbz(vcpu, &pte); 350 } else { 351 /* MMIO */ 352 vcpu->stat.mmio_exits++; 353 vcpu->arch.paddr_accessed = pte.raddr; 354 r = kvmppc_emulate_mmio(run, vcpu); 355 if ( r == RESUME_HOST_NV ) 356 r = RESUME_HOST; 357 } 358 359 return r; 360 } 361 362 static inline int get_fpr_index(int i) 363 { 364 #ifdef CONFIG_VSX 365 i *= 2; 366 #endif 367 return i; 368 } 369 370 /* Give up external provider (FPU, Altivec, VSX) */ 371 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr) 372 { 373 struct thread_struct *t = ¤t->thread; 374 u64 *vcpu_fpr = vcpu->arch.fpr; 375 #ifdef CONFIG_VSX 376 u64 *vcpu_vsx = vcpu->arch.vsr; 377 #endif 378 u64 *thread_fpr = (u64*)t->fpr; 379 int i; 380 381 if (!(vcpu->arch.guest_owned_ext & msr)) 382 return; 383 384 #ifdef DEBUG_EXT 385 printk(KERN_INFO "Giving up ext 0x%lx\n", msr); 386 #endif 387 388 switch (msr) { 389 case MSR_FP: 390 giveup_fpu(current); 391 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++) 392 vcpu_fpr[i] = thread_fpr[get_fpr_index(i)]; 393 394 vcpu->arch.fpscr = t->fpscr.val; 395 break; 396 case MSR_VEC: 397 #ifdef CONFIG_ALTIVEC 398 giveup_altivec(current); 399 memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr)); 400 vcpu->arch.vscr = t->vscr; 401 #endif 402 break; 403 case MSR_VSX: 404 #ifdef CONFIG_VSX 405 __giveup_vsx(current); 406 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++) 407 vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1]; 408 #endif 409 break; 410 default: 411 BUG(); 412 } 413 414 vcpu->arch.guest_owned_ext &= ~msr; 415 current->thread.regs->msr &= ~msr; 416 kvmppc_recalc_shadow_msr(vcpu); 417 } 418 419 static int kvmppc_read_inst(struct kvm_vcpu *vcpu) 420 { 421 ulong srr0 = kvmppc_get_pc(vcpu); 422 u32 last_inst = kvmppc_get_last_inst(vcpu); 423 int ret; 424 425 ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false); 426 if (ret == -ENOENT) { 427 ulong msr = vcpu->arch.shared->msr; 428 429 msr = kvmppc_set_field(msr, 33, 33, 1); 430 msr = kvmppc_set_field(msr, 34, 36, 0); 431 vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0); 432 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE); 433 return EMULATE_AGAIN; 434 } 435 436 return EMULATE_DONE; 437 } 438 439 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr) 440 { 441 442 /* Need to do paired single emulation? */ 443 if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)) 444 return EMULATE_DONE; 445 446 /* Read out the instruction */ 447 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) 448 /* Need to emulate */ 449 return EMULATE_FAIL; 450 451 return EMULATE_AGAIN; 452 } 453 454 /* Handle external providers (FPU, Altivec, VSX) */ 455 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr, 456 ulong msr) 457 { 458 struct thread_struct *t = ¤t->thread; 459 u64 *vcpu_fpr = vcpu->arch.fpr; 460 #ifdef CONFIG_VSX 461 u64 *vcpu_vsx = vcpu->arch.vsr; 462 #endif 463 u64 *thread_fpr = (u64*)t->fpr; 464 int i; 465 466 /* When we have paired singles, we emulate in software */ 467 if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE) 468 return RESUME_GUEST; 469 470 if (!(vcpu->arch.shared->msr & msr)) { 471 kvmppc_book3s_queue_irqprio(vcpu, exit_nr); 472 return RESUME_GUEST; 473 } 474 475 /* We already own the ext */ 476 if (vcpu->arch.guest_owned_ext & msr) { 477 return RESUME_GUEST; 478 } 479 480 #ifdef DEBUG_EXT 481 printk(KERN_INFO "Loading up ext 0x%lx\n", msr); 482 #endif 483 484 current->thread.regs->msr |= msr; 485 486 switch (msr) { 487 case MSR_FP: 488 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++) 489 thread_fpr[get_fpr_index(i)] = vcpu_fpr[i]; 490 491 t->fpscr.val = vcpu->arch.fpscr; 492 t->fpexc_mode = 0; 493 kvmppc_load_up_fpu(); 494 break; 495 case MSR_VEC: 496 #ifdef CONFIG_ALTIVEC 497 memcpy(t->vr, vcpu->arch.vr, sizeof(vcpu->arch.vr)); 498 t->vscr = vcpu->arch.vscr; 499 t->vrsave = -1; 500 kvmppc_load_up_altivec(); 501 #endif 502 break; 503 case MSR_VSX: 504 #ifdef CONFIG_VSX 505 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++) 506 thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i]; 507 kvmppc_load_up_vsx(); 508 #endif 509 break; 510 default: 511 BUG(); 512 } 513 514 vcpu->arch.guest_owned_ext |= msr; 515 516 kvmppc_recalc_shadow_msr(vcpu); 517 518 return RESUME_GUEST; 519 } 520 521 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu, 522 unsigned int exit_nr) 523 { 524 int r = RESUME_HOST; 525 526 vcpu->stat.sum_exits++; 527 528 run->exit_reason = KVM_EXIT_UNKNOWN; 529 run->ready_for_interrupt_injection = 1; 530 531 trace_kvm_book3s_exit(exit_nr, vcpu); 532 preempt_enable(); 533 kvm_resched(vcpu); 534 switch (exit_nr) { 535 case BOOK3S_INTERRUPT_INST_STORAGE: 536 { 537 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu); 538 ulong shadow_srr1 = svcpu->shadow_srr1; 539 vcpu->stat.pf_instruc++; 540 541 #ifdef CONFIG_PPC_BOOK3S_32 542 /* We set segments as unused segments when invalidating them. So 543 * treat the respective fault as segment fault. */ 544 if (svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT] == SR_INVALID) { 545 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)); 546 r = RESUME_GUEST; 547 svcpu_put(svcpu); 548 break; 549 } 550 #endif 551 svcpu_put(svcpu); 552 553 /* only care about PTEG not found errors, but leave NX alone */ 554 if (shadow_srr1 & 0x40000000) { 555 r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr); 556 vcpu->stat.sp_instruc++; 557 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) && 558 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) { 559 /* 560 * XXX If we do the dcbz hack we use the NX bit to flush&patch the page, 561 * so we can't use the NX bit inside the guest. Let's cross our fingers, 562 * that no guest that needs the dcbz hack does NX. 563 */ 564 kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL); 565 r = RESUME_GUEST; 566 } else { 567 vcpu->arch.shared->msr |= shadow_srr1 & 0x58000000; 568 kvmppc_book3s_queue_irqprio(vcpu, exit_nr); 569 r = RESUME_GUEST; 570 } 571 break; 572 } 573 case BOOK3S_INTERRUPT_DATA_STORAGE: 574 { 575 ulong dar = kvmppc_get_fault_dar(vcpu); 576 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu); 577 u32 fault_dsisr = svcpu->fault_dsisr; 578 vcpu->stat.pf_storage++; 579 580 #ifdef CONFIG_PPC_BOOK3S_32 581 /* We set segments as unused segments when invalidating them. So 582 * treat the respective fault as segment fault. */ 583 if ((svcpu->sr[dar >> SID_SHIFT]) == SR_INVALID) { 584 kvmppc_mmu_map_segment(vcpu, dar); 585 r = RESUME_GUEST; 586 svcpu_put(svcpu); 587 break; 588 } 589 #endif 590 svcpu_put(svcpu); 591 592 /* The only case we need to handle is missing shadow PTEs */ 593 if (fault_dsisr & DSISR_NOHPTE) { 594 r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr); 595 } else { 596 vcpu->arch.shared->dar = dar; 597 vcpu->arch.shared->dsisr = fault_dsisr; 598 kvmppc_book3s_queue_irqprio(vcpu, exit_nr); 599 r = RESUME_GUEST; 600 } 601 break; 602 } 603 case BOOK3S_INTERRUPT_DATA_SEGMENT: 604 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) { 605 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu); 606 kvmppc_book3s_queue_irqprio(vcpu, 607 BOOK3S_INTERRUPT_DATA_SEGMENT); 608 } 609 r = RESUME_GUEST; 610 break; 611 case BOOK3S_INTERRUPT_INST_SEGMENT: 612 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) { 613 kvmppc_book3s_queue_irqprio(vcpu, 614 BOOK3S_INTERRUPT_INST_SEGMENT); 615 } 616 r = RESUME_GUEST; 617 break; 618 /* We're good on these - the host merely wanted to get our attention */ 619 case BOOK3S_INTERRUPT_DECREMENTER: 620 vcpu->stat.dec_exits++; 621 r = RESUME_GUEST; 622 break; 623 case BOOK3S_INTERRUPT_EXTERNAL: 624 vcpu->stat.ext_intr_exits++; 625 r = RESUME_GUEST; 626 break; 627 case BOOK3S_INTERRUPT_PERFMON: 628 r = RESUME_GUEST; 629 break; 630 case BOOK3S_INTERRUPT_PROGRAM: 631 { 632 enum emulation_result er; 633 struct kvmppc_book3s_shadow_vcpu *svcpu; 634 ulong flags; 635 636 program_interrupt: 637 svcpu = svcpu_get(vcpu); 638 flags = svcpu->shadow_srr1 & 0x1f0000ull; 639 svcpu_put(svcpu); 640 641 if (vcpu->arch.shared->msr & MSR_PR) { 642 #ifdef EXIT_DEBUG 643 printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu)); 644 #endif 645 if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) != 646 (INS_DCBZ & 0xfffffff7)) { 647 kvmppc_core_queue_program(vcpu, flags); 648 r = RESUME_GUEST; 649 break; 650 } 651 } 652 653 vcpu->stat.emulated_inst_exits++; 654 er = kvmppc_emulate_instruction(run, vcpu); 655 switch (er) { 656 case EMULATE_DONE: 657 r = RESUME_GUEST_NV; 658 break; 659 case EMULATE_AGAIN: 660 r = RESUME_GUEST; 661 break; 662 case EMULATE_FAIL: 663 printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n", 664 __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu)); 665 kvmppc_core_queue_program(vcpu, flags); 666 r = RESUME_GUEST; 667 break; 668 case EMULATE_DO_MMIO: 669 run->exit_reason = KVM_EXIT_MMIO; 670 r = RESUME_HOST_NV; 671 break; 672 default: 673 BUG(); 674 } 675 break; 676 } 677 case BOOK3S_INTERRUPT_SYSCALL: 678 if (vcpu->arch.papr_enabled && 679 (kvmppc_get_last_inst(vcpu) == 0x44000022) && 680 !(vcpu->arch.shared->msr & MSR_PR)) { 681 /* SC 1 papr hypercalls */ 682 ulong cmd = kvmppc_get_gpr(vcpu, 3); 683 int i; 684 685 #ifdef CONFIG_KVM_BOOK3S_64_PR 686 if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) { 687 r = RESUME_GUEST; 688 break; 689 } 690 #endif 691 692 run->papr_hcall.nr = cmd; 693 for (i = 0; i < 9; ++i) { 694 ulong gpr = kvmppc_get_gpr(vcpu, 4 + i); 695 run->papr_hcall.args[i] = gpr; 696 } 697 run->exit_reason = KVM_EXIT_PAPR_HCALL; 698 vcpu->arch.hcall_needed = 1; 699 r = RESUME_HOST; 700 } else if (vcpu->arch.osi_enabled && 701 (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) && 702 (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) { 703 /* MOL hypercalls */ 704 u64 *gprs = run->osi.gprs; 705 int i; 706 707 run->exit_reason = KVM_EXIT_OSI; 708 for (i = 0; i < 32; i++) 709 gprs[i] = kvmppc_get_gpr(vcpu, i); 710 vcpu->arch.osi_needed = 1; 711 r = RESUME_HOST_NV; 712 } else if (!(vcpu->arch.shared->msr & MSR_PR) && 713 (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) { 714 /* KVM PV hypercalls */ 715 kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu)); 716 r = RESUME_GUEST; 717 } else { 718 /* Guest syscalls */ 719 vcpu->stat.syscall_exits++; 720 kvmppc_book3s_queue_irqprio(vcpu, exit_nr); 721 r = RESUME_GUEST; 722 } 723 break; 724 case BOOK3S_INTERRUPT_FP_UNAVAIL: 725 case BOOK3S_INTERRUPT_ALTIVEC: 726 case BOOK3S_INTERRUPT_VSX: 727 { 728 int ext_msr = 0; 729 730 switch (exit_nr) { 731 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP; break; 732 case BOOK3S_INTERRUPT_ALTIVEC: ext_msr = MSR_VEC; break; 733 case BOOK3S_INTERRUPT_VSX: ext_msr = MSR_VSX; break; 734 } 735 736 switch (kvmppc_check_ext(vcpu, exit_nr)) { 737 case EMULATE_DONE: 738 /* everything ok - let's enable the ext */ 739 r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr); 740 break; 741 case EMULATE_FAIL: 742 /* we need to emulate this instruction */ 743 goto program_interrupt; 744 break; 745 default: 746 /* nothing to worry about - go again */ 747 break; 748 } 749 break; 750 } 751 case BOOK3S_INTERRUPT_ALIGNMENT: 752 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) { 753 vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu, 754 kvmppc_get_last_inst(vcpu)); 755 vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu, 756 kvmppc_get_last_inst(vcpu)); 757 kvmppc_book3s_queue_irqprio(vcpu, exit_nr); 758 } 759 r = RESUME_GUEST; 760 break; 761 case BOOK3S_INTERRUPT_MACHINE_CHECK: 762 case BOOK3S_INTERRUPT_TRACE: 763 kvmppc_book3s_queue_irqprio(vcpu, exit_nr); 764 r = RESUME_GUEST; 765 break; 766 default: 767 { 768 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu); 769 ulong shadow_srr1 = svcpu->shadow_srr1; 770 svcpu_put(svcpu); 771 /* Ugh - bork here! What did we get? */ 772 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n", 773 exit_nr, kvmppc_get_pc(vcpu), shadow_srr1); 774 r = RESUME_HOST; 775 BUG(); 776 break; 777 } 778 } 779 780 preempt_disable(); 781 if (!(r & RESUME_HOST)) { 782 /* To avoid clobbering exit_reason, only check for signals if 783 * we aren't already exiting to userspace for some other 784 * reason. */ 785 786 /* 787 * Interrupts could be timers for the guest which we have to 788 * inject again, so let's postpone them until we're in the guest 789 * and if we really did time things so badly, then we just exit 790 * again due to a host external interrupt. 791 */ 792 __hard_irq_disable(); 793 if (signal_pending(current)) { 794 __hard_irq_enable(); 795 #ifdef EXIT_DEBUG 796 printk(KERN_EMERG "KVM: Going back to host\n"); 797 #endif 798 vcpu->stat.signal_exits++; 799 run->exit_reason = KVM_EXIT_INTR; 800 r = -EINTR; 801 } else { 802 /* In case an interrupt came in that was triggered 803 * from userspace (like DEC), we need to check what 804 * to inject now! */ 805 kvmppc_core_prepare_to_enter(vcpu); 806 } 807 } 808 809 trace_kvm_book3s_reenter(r, vcpu); 810 811 return r; 812 } 813 814 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu, 815 struct kvm_sregs *sregs) 816 { 817 struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu); 818 int i; 819 820 sregs->pvr = vcpu->arch.pvr; 821 822 sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1; 823 if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) { 824 for (i = 0; i < 64; i++) { 825 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i; 826 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv; 827 } 828 } else { 829 for (i = 0; i < 16; i++) 830 sregs->u.s.ppc32.sr[i] = vcpu->arch.shared->sr[i]; 831 832 for (i = 0; i < 8; i++) { 833 sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw; 834 sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw; 835 } 836 } 837 838 return 0; 839 } 840 841 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu, 842 struct kvm_sregs *sregs) 843 { 844 struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu); 845 int i; 846 847 kvmppc_set_pvr(vcpu, sregs->pvr); 848 849 vcpu3s->sdr1 = sregs->u.s.sdr1; 850 if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) { 851 for (i = 0; i < 64; i++) { 852 vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv, 853 sregs->u.s.ppc64.slb[i].slbe); 854 } 855 } else { 856 for (i = 0; i < 16; i++) { 857 vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]); 858 } 859 for (i = 0; i < 8; i++) { 860 kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false, 861 (u32)sregs->u.s.ppc32.ibat[i]); 862 kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true, 863 (u32)(sregs->u.s.ppc32.ibat[i] >> 32)); 864 kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false, 865 (u32)sregs->u.s.ppc32.dbat[i]); 866 kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true, 867 (u32)(sregs->u.s.ppc32.dbat[i] >> 32)); 868 } 869 } 870 871 /* Flush the MMU after messing with the segments */ 872 kvmppc_mmu_pte_flush(vcpu, 0, 0); 873 874 return 0; 875 } 876 877 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg) 878 { 879 int r = -EINVAL; 880 881 switch (reg->id) { 882 case KVM_REG_PPC_HIOR: 883 r = copy_to_user((u64 __user *)(long)reg->addr, 884 &to_book3s(vcpu)->hior, sizeof(u64)); 885 break; 886 default: 887 break; 888 } 889 890 return r; 891 } 892 893 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg) 894 { 895 int r = -EINVAL; 896 897 switch (reg->id) { 898 case KVM_REG_PPC_HIOR: 899 r = copy_from_user(&to_book3s(vcpu)->hior, 900 (u64 __user *)(long)reg->addr, sizeof(u64)); 901 if (!r) 902 to_book3s(vcpu)->hior_explicit = true; 903 break; 904 default: 905 break; 906 } 907 908 return r; 909 } 910 911 int kvmppc_core_check_processor_compat(void) 912 { 913 return 0; 914 } 915 916 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id) 917 { 918 struct kvmppc_vcpu_book3s *vcpu_book3s; 919 struct kvm_vcpu *vcpu; 920 int err = -ENOMEM; 921 unsigned long p; 922 923 vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s)); 924 if (!vcpu_book3s) 925 goto out; 926 927 vcpu_book3s->shadow_vcpu = (struct kvmppc_book3s_shadow_vcpu *) 928 kzalloc(sizeof(*vcpu_book3s->shadow_vcpu), GFP_KERNEL); 929 if (!vcpu_book3s->shadow_vcpu) 930 goto free_vcpu; 931 932 vcpu = &vcpu_book3s->vcpu; 933 err = kvm_vcpu_init(vcpu, kvm, id); 934 if (err) 935 goto free_shadow_vcpu; 936 937 p = __get_free_page(GFP_KERNEL|__GFP_ZERO); 938 /* the real shared page fills the last 4k of our page */ 939 vcpu->arch.shared = (void*)(p + PAGE_SIZE - 4096); 940 if (!p) 941 goto uninit_vcpu; 942 943 #ifdef CONFIG_PPC_BOOK3S_64 944 /* default to book3s_64 (970fx) */ 945 vcpu->arch.pvr = 0x3C0301; 946 #else 947 /* default to book3s_32 (750) */ 948 vcpu->arch.pvr = 0x84202; 949 #endif 950 kvmppc_set_pvr(vcpu, vcpu->arch.pvr); 951 vcpu->arch.slb_nr = 64; 952 953 vcpu->arch.shadow_msr = MSR_USER64; 954 955 err = kvmppc_mmu_init(vcpu); 956 if (err < 0) 957 goto uninit_vcpu; 958 959 return vcpu; 960 961 uninit_vcpu: 962 kvm_vcpu_uninit(vcpu); 963 free_shadow_vcpu: 964 kfree(vcpu_book3s->shadow_vcpu); 965 free_vcpu: 966 vfree(vcpu_book3s); 967 out: 968 return ERR_PTR(err); 969 } 970 971 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu) 972 { 973 struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu); 974 975 free_page((unsigned long)vcpu->arch.shared & PAGE_MASK); 976 kvm_vcpu_uninit(vcpu); 977 kfree(vcpu_book3s->shadow_vcpu); 978 vfree(vcpu_book3s); 979 } 980 981 int kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu) 982 { 983 int ret; 984 double fpr[32][TS_FPRWIDTH]; 985 unsigned int fpscr; 986 int fpexc_mode; 987 #ifdef CONFIG_ALTIVEC 988 vector128 vr[32]; 989 vector128 vscr; 990 unsigned long uninitialized_var(vrsave); 991 int used_vr; 992 #endif 993 #ifdef CONFIG_VSX 994 int used_vsr; 995 #endif 996 ulong ext_msr; 997 998 preempt_disable(); 999 1000 /* Check if we can run the vcpu at all */ 1001 if (!vcpu->arch.sane) { 1002 kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR; 1003 ret = -EINVAL; 1004 goto out; 1005 } 1006 1007 kvmppc_core_prepare_to_enter(vcpu); 1008 1009 /* 1010 * Interrupts could be timers for the guest which we have to inject 1011 * again, so let's postpone them until we're in the guest and if we 1012 * really did time things so badly, then we just exit again due to 1013 * a host external interrupt. 1014 */ 1015 __hard_irq_disable(); 1016 1017 /* No need to go into the guest when all we do is going out */ 1018 if (signal_pending(current)) { 1019 __hard_irq_enable(); 1020 kvm_run->exit_reason = KVM_EXIT_INTR; 1021 ret = -EINTR; 1022 goto out; 1023 } 1024 1025 /* Save FPU state in stack */ 1026 if (current->thread.regs->msr & MSR_FP) 1027 giveup_fpu(current); 1028 memcpy(fpr, current->thread.fpr, sizeof(current->thread.fpr)); 1029 fpscr = current->thread.fpscr.val; 1030 fpexc_mode = current->thread.fpexc_mode; 1031 1032 #ifdef CONFIG_ALTIVEC 1033 /* Save Altivec state in stack */ 1034 used_vr = current->thread.used_vr; 1035 if (used_vr) { 1036 if (current->thread.regs->msr & MSR_VEC) 1037 giveup_altivec(current); 1038 memcpy(vr, current->thread.vr, sizeof(current->thread.vr)); 1039 vscr = current->thread.vscr; 1040 vrsave = current->thread.vrsave; 1041 } 1042 #endif 1043 1044 #ifdef CONFIG_VSX 1045 /* Save VSX state in stack */ 1046 used_vsr = current->thread.used_vsr; 1047 if (used_vsr && (current->thread.regs->msr & MSR_VSX)) 1048 __giveup_vsx(current); 1049 #endif 1050 1051 /* Remember the MSR with disabled extensions */ 1052 ext_msr = current->thread.regs->msr; 1053 1054 /* Preload FPU if it's enabled */ 1055 if (vcpu->arch.shared->msr & MSR_FP) 1056 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP); 1057 1058 kvm_guest_enter(); 1059 1060 ret = __kvmppc_vcpu_run(kvm_run, vcpu); 1061 1062 kvm_guest_exit(); 1063 1064 current->thread.regs->msr = ext_msr; 1065 1066 /* Make sure we save the guest FPU/Altivec/VSX state */ 1067 kvmppc_giveup_ext(vcpu, MSR_FP); 1068 kvmppc_giveup_ext(vcpu, MSR_VEC); 1069 kvmppc_giveup_ext(vcpu, MSR_VSX); 1070 1071 /* Restore FPU state from stack */ 1072 memcpy(current->thread.fpr, fpr, sizeof(current->thread.fpr)); 1073 current->thread.fpscr.val = fpscr; 1074 current->thread.fpexc_mode = fpexc_mode; 1075 1076 #ifdef CONFIG_ALTIVEC 1077 /* Restore Altivec state from stack */ 1078 if (used_vr && current->thread.used_vr) { 1079 memcpy(current->thread.vr, vr, sizeof(current->thread.vr)); 1080 current->thread.vscr = vscr; 1081 current->thread.vrsave = vrsave; 1082 } 1083 current->thread.used_vr = used_vr; 1084 #endif 1085 1086 #ifdef CONFIG_VSX 1087 current->thread.used_vsr = used_vsr; 1088 #endif 1089 1090 out: 1091 preempt_enable(); 1092 return ret; 1093 } 1094 1095 /* 1096 * Get (and clear) the dirty memory log for a memory slot. 1097 */ 1098 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, 1099 struct kvm_dirty_log *log) 1100 { 1101 struct kvm_memory_slot *memslot; 1102 struct kvm_vcpu *vcpu; 1103 ulong ga, ga_end; 1104 int is_dirty = 0; 1105 int r; 1106 unsigned long n; 1107 1108 mutex_lock(&kvm->slots_lock); 1109 1110 r = kvm_get_dirty_log(kvm, log, &is_dirty); 1111 if (r) 1112 goto out; 1113 1114 /* If nothing is dirty, don't bother messing with page tables. */ 1115 if (is_dirty) { 1116 memslot = id_to_memslot(kvm->memslots, log->slot); 1117 1118 ga = memslot->base_gfn << PAGE_SHIFT; 1119 ga_end = ga + (memslot->npages << PAGE_SHIFT); 1120 1121 kvm_for_each_vcpu(n, vcpu, kvm) 1122 kvmppc_mmu_pte_pflush(vcpu, ga, ga_end); 1123 1124 n = kvm_dirty_bitmap_bytes(memslot); 1125 memset(memslot->dirty_bitmap, 0, n); 1126 } 1127 1128 r = 0; 1129 out: 1130 mutex_unlock(&kvm->slots_lock); 1131 return r; 1132 } 1133 1134 int kvmppc_core_prepare_memory_region(struct kvm *kvm, 1135 struct kvm_userspace_memory_region *mem) 1136 { 1137 return 0; 1138 } 1139 1140 void kvmppc_core_commit_memory_region(struct kvm *kvm, 1141 struct kvm_userspace_memory_region *mem) 1142 { 1143 } 1144 1145 int kvmppc_core_init_vm(struct kvm *kvm) 1146 { 1147 return 0; 1148 } 1149 1150 void kvmppc_core_destroy_vm(struct kvm *kvm) 1151 { 1152 } 1153 1154 static int kvmppc_book3s_init(void) 1155 { 1156 int r; 1157 1158 r = kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), 0, 1159 THIS_MODULE); 1160 1161 if (r) 1162 return r; 1163 1164 r = kvmppc_mmu_hpte_sysinit(); 1165 1166 return r; 1167 } 1168 1169 static void kvmppc_book3s_exit(void) 1170 { 1171 kvmppc_mmu_hpte_sysexit(); 1172 kvm_exit(); 1173 } 1174 1175 module_init(kvmppc_book3s_init); 1176 module_exit(kvmppc_book3s_exit); 1177